Geometric information extraction for the camera cooperation without positions information

K. Takano, T. Suzuki, Hajime Asama, Kenichi MishimaThe Proceedings of JSME Annual Conference on Robotics and Mechatronics (Robomec) 2002 · DOI

Abstract

In cooperative vision systems, establishing the spatial relationship between multiple cameras is a fundamental requirement. However, traditional calibration methods often fail or become impractical when camera positions are unknown or change dynamically. This paper proposes a method to extract geometric information between multiple active cameras using a laser pointer as a shared reference. By tracking the projected laser spot, the system can determine the necessary mapping between image planes without prior knowledge of the cameras' extrinsic parameters.

Introduction

Cooperative vision systems—where multiple cameras work in tandem to track objects or map environments—rely heavily on "extrinsic parameters," which define the relative position and orientation of each camera. In static environments, these are typically measured manually or through checkerboard calibration. However, for active cameras (those capable of panning and tilting) or cameras mounted on mobile robots, maintaining an accurate spatial model is a significant challenge.

This research addresses the problem of establishing a geometric link between cameras when their spatial arrangement is entirely unknown. The goal is to enable seamless cooperation without the overhead of traditional calibration routines.

Methodology: The Laser as a Universal Reference

The proposed system utilizes two primary components: active pan-tilt cameras and a laser pointer. The process follows a structured workflow:

  1. Projection of a Virtual Feature: Instead of relying on environmental landmarks, which may be sparse or ambiguous, a laser pointer projects a high-contrast spot onto a surface within the shared field of view.
  2. Synchronized Spot Detection: Because the laser spot is a distinct, high-intensity feature, it is easily identified by all cameras. This creates a "correspondence" point—a single physical location represented as $(u, v)$ coordinates in each camera's image plane.
  3. Active Tracking and Data Collection: To build a robust geometric model, the cameras "actively" track the laser spot as it moves. This movement allows the system to collect a diverse set of corresponding points across the cameras' fields of view.
  4. Geometric Mapping: Using these correspondences, the system calculates the geometric relationship (such as a fundamental matrix or homography) required to map coordinates from one camera's view to another.

Key Insights

The "Virtual Feature" Concept

The most significant contribution of this work is the use of a laser to create a "perfect" feature. In computer vision, the "correspondence problem"—determining if Point A in Camera 1 is the same as Point B in Camera 2—is notoriously difficult. A laser spot eliminates this ambiguity, providing a clear, high-confidence signal for calibration.

Decoupling Geometry from Physical Position

A crucial realization in this research is that for many cooperative tasks, the exact XYZ coordinates of the cameras are not strictly necessary. If the system can mathematically translate the pixel coordinates of one camera to those of another, it has sufficient geometric information to perform tracking and hand-over tasks.

Environmental Independence

Because the system generates its own reference point, it is highly robust. It can function in featureless environments, such as empty warehouses or rooms with repetitive patterns, where traditional feature-matching algorithms (like SIFT or ORB) would likely fail.

Conclusion and Future Work

The research successfully demonstrates that geometric information can be extracted from laser spot projections even when camera positions are unknown. This provides a flexible foundation for ad-hoc cooperative vision systems.

Future developments may focus on the challenges of line-of-sight—ensuring that the laser remains visible to all cameras—and optimizing the system for real-time performance on mobile platforms where camera positions are constantly in flux.

Summary updated

BibTeX
@inproceedings{takano2002geometric,
  author = {K. Takano and T. Suzuki and Hajime Asama and Kenichi Mishima},
  title = {{Geometric information extraction for the camera cooperation without positions information}},
  booktitle = {The Proceedings of JSME Annual Conference on Robotics and Mechatronics (Robomec)},
  year = {2002},
  doi = {10.1299/jsmermd.2002.119\_5},
}
Figure